Learning mechanism of chromatin domain formation with big data

Chromatin modifications play critical roles in gene regulation and encoding cell phenotypic diversity. The molecular mechanism for their establishment and maintenance is not fully understood due to the complexity of chromatin regulatory pathways. Here we took a data-driven approach and parameterized an information-theoretic model to infer mechanism of chromatin domain formation from genome-wide epigenetic modification profiles. The energy landscape of this model reveals many important chromatin domains that span multiple nucleosomes and exhibit distinct combinatorial patterns of histone modifications, including super (stretch) enhancers, broad H3K4me3 promoter domains, heterochromatin, etc. Transition path analysis further demonstrates that enhancer and promoter domains undergo a sequential maturation process along which the regulatory elements grow from short but stable nucleosome segments to long and potent ones that are modified with many activation marks. On the other hand, the formation of heterochromatin domains is a highly cooperative process, and no intermediate states were found along the transition path. Interaction energies of the information-theoretic model further suggest that heterochromatin domains adopt collapsed, globular three-dimensional conformations that can be stabilized by phase-separated liquid droplets.Our results demonstrate the usefulness of statistical mechanical models and molecular biophysical approaches in interpreting the rich information encoded in epigenomics data.